Downloads · 30 days
16
17% of all-time downloads
Kamciosz/qwen38-mtp-head-xkm-affine2-v1
qwen38-mtp-head-xkm-affine2-v1 is a machine learning model from Kamciosz. Use it for the machine learning task on the model card, and read the license before you ship it in a product. It is set up for mlx. The card lists the license as apache-2.0.
Proposal-only head for the Qwen 3.8 native-MTP decode track.
Downloads · 30 days
16
17% of all-time downloads
All-time downloads
92
Public
Parameters
472M
1 GB on disk
Likes
0
Public
Click a slice to open those files.
.safetensors1 GB · 100%
How the weights are stored.
BF16440M · 93%
From the Hugging Face model README
Proposal-only head for the Qwen 3.8 native-MTP decode track.
xkm/qwen3.8-27b-mtp-head-retrained
(chain-faithful distillation vs the target argmax).draft_lm_head.{weight,scales,biases}: byte copies from
amal-david/qwen38-mtp-head-q2-q4-rerank-v1 (affine-2 compact shortlist).config.json: compatibility copy of EigenLabs/Qwen3.8-27B-MTP-bf16
(qwen3_5_mtp, mtp_num_hidden_layers: 1).The target backbone, target top-2, accept walk, and serial control are unchanged. This artifact only proposes tokens.
Parents retain their licenses (Apache-2.0). No new training was run for this merge.